democracy and civil war - · pdf filedemocracy and civil war1 nils petter gleditsch,...
TRANSCRIPT
Democracy and Civil War1
Nils Petter Gleditsch, Håvard Hegre, and Håvard Strand2
Democracy relates closely to civil war. In this chapter, we investigate ten propositions: De-mocracy is negatively related to the onset and severity of civil war and positively to dura-tion. Democracy relates to civil war through an inverted U-curve: semi-democracies are most prone to violence. Political instability is positively related to civil war. The curvilinear effect of democracy on civil war persists even when controlling for political instability. Civil war occurs more frequently in conjunction with elections. The effect of democracy on civil war is more pronounced for government conflicts than for territorial conflicts. The effect of democracy on civil war is stronger for developed countries. A politically different neighbor-hood is positively related to civil war. The negative effect of democracy on civil war is more pronounced for inclusive types of democracy. These relationships are generally strength-ened after the end of the Cold War. Using two different measures of democracy, we test these hypotheses on the Uppsala/PRIO conflict data. Overall, we find democracy to be strongly related to civil war, with the results for severity being the most robust.
For Manus Midlarsky, ed., Handbook of War Studies III (Ann Arbor, MI: Uni-versity of Michigan Press, 2007)
NILS PETTER GLEDITSCH, Research professor, Centre for the Study of Civil War (CSCW), International Peace Research Institute (PRIO) and Professor of poli tical science, Norwegian University of Science and Technology; [email protected] HÅVARD HEGRE, Research professor, Centre for the Stu dy of Civil War, International Peace Research Institute, Oslo (PRIO); [email protected] HÅVARD STRAND, PhD in political science, University o f Oslo; Research fellow, Centre for the Study of Civil War, International Peace Research Institute, Oslo (PRIO); [email protected]
1 Our work on this article was supported by the Research Council of Norway. An earlier version was presented at a workshop on ‘Polarization and Conflict’, Nicosia, Cyprus, 26–29 April 2006. We are grateful to the participants of that meeting, Lars Wilhelmsen, and the editor of this volume for com-ments and to Siri Aas Rustad for creating the map in Figure 1. Replication data for the analyses per-formed here can be found on www.prio.no/cscw/datasets along with an appendix describing the statistical tests. 2 Alphabetical order; equal co-authorship implied.
2
Introduction
In this chapter, we investigate the question of the democratic civil peace – i.e. the
democratic peace at the intrastate level. We use interchangeably the terms civil war
and intrastate violence for events where organized violence is used for political
goals, although conventionally the term “war” is often reserved for conflicts where
the annual number of battle deaths exceed 1,000. We study the determinants of
the onset of civil war as well as the incidence, duration, and severity of violent
events.
Recent research on civil war uses arguments regarding the opportunity and
motivation for rebellion as a theoretical point of departure. We discuss how demo-
cratic institutions may affect both the opportunity and the willingness to use vio-
lence for political purposes. Democratic institutions display great variations,
however, in terms of inclusiveness, effectiveness, and stability. The opportunity and
motivation framework allows us to derive a set of expectations concerning when
and under what conditions democratic institutions may succeed in reducing the
risk, incidence, duration, and severity of civil war.
The general hypothesis emerging from this discussion is that democratic gov-
ernance is beneficial for the reduction of civil war and that specific forms of democ-
racy are likely to reduce violence more than others. Specifically, we investigate
these propositions: Democracy is negatively related to the onset and severity of civil
war and positively to duration. Democracy relates to civil war through an inverted
U-curve: semi-democracies are most prone to violence. Political instability is posi-
tively related to civil war. The curvilinear effect of democracy on civil war persists
even when controlling for political instability. Civil war occurs more frequently in
conjunction with elections. The effect of democracy on civil war is more pronounced
for government conflicts than for territorial conflicts. The effect of democracy on
civil war is stronger for developed countries. A politically different neighborhood is
positively related to civil war. The negative effect of democracy on civil war is more
pronounced for inclusive types of democracy. These relationships are generally
strengthened after the end of the Cold War. We test these hypotheses on the Upp-
3
sala/PRIO conflict data, and investigate the robustness of our results. We use two
measures of democracy, the democracy index from the Polity project and the new
SIP index, with some robustness tests using two additional measures, Vanhanen’s
Polyarchy index and the Freedom House Political Rights measure. Overall, we find
democracy to be strongly related to civil war, with the results for severity being the
most robust.
Theoretical framework
Actors All organized violence is by definition a dyadic phenomenon, where someone acts
violently upon someone else. Two actors are required, both organized parties, one
of which is a government of an independent country.3 Our main interest lies in the
characteristics of the government side. Additional actors may also be involved, in-
cluding other governments, but here we assume that they are allies of one or the
other of the primary actors. In the analyses reported in this chapter, we use no da-
ta about the opposition side, except that it is organized.4 This is not necessarily a
serious limitation, since our interest is in exploring the effects of the system of gov-
ernance on civil war. However, in future extensions it might be of interest to look at
the political ideology and aspirations of the opposition side for a more truly dyadic
analysis, comparable to the analysis of the dyadic democratic peace at the inter-
state level.
Much scholarly debate has revolved around what type of actor a typical rebel
group is: Is it the military organization of a broad social movement that seeks to
address government injustice or promote democratization (e.g. Lichbach 1995), or
the armed branch of a minority group that requires some sort of cultural autonomy
relative to the government or the country’s majority (Sambanis 2001), or is it just a
3 In the Uppsala/PRIO terminology (see Gleditsch et al. 2002, Strand et al. 2005), which we generally follow, conflicts where no government is involved (sometimes labeled communal conflicts) are termed non-state conflicts. Conflicts where the nonstate actor is unorganized (variously genocide, politicide, or democide) are labeled one-sided conflicts. See Mack (2006) for a presentation of data on non-state conflicts 2002–05 and Eck and Hultman (2007) for a presentation of data on one-sided conflicts 1989–2004. 4 For dyadic analyses of civil war with more data on the opposition side, see Gleditsch, Cunningham, and Salehyan (2006) and Raleigh and Hegre (2006).
4
military organization designed to forward the private political and economic inter-
ests of its leaders (Collier 2000a)? Actors may also be a combination of these types,
and transform from one to another in the course of the conflict. Examples of all
these types may be found among the world’s rebel organizations, but the empirical
study of civil war is affected by how common each of these are. Rebel groups are
not homogenous either, and the leaders’ motivations may be very different from
those of ordinary soldiers.
Motivation and opportunity We conceptualize civil war as a product of motive and opportunity. On the rebel
side, a motive can be negative – a grievance against the existing state of affairs, or
positive – a desire to get rich, sometimes called greed in the civil war literature. Re-
searchers that focus on grievance typically discuss broad social movements,
whereas greed motivations tend to be associated with narrow groups or elites with-
in such movements. A prominent example of the first category is the ANC’s decision
to take up arms against the apartheid regime, while the latter category is often il-
lustrated with Charles Taylor’s rise (and fall) in Liberia.
Second, potential rebels must have a realistic opportunity to achieve their
goal. The realization of their desires may be blocked by a powerful well-organized
government army, by geographical factors (if the rebels are thinly spread out or are
unable to establish defensible headquarters), by the lack of financial means (if they
are very poor), or by the inability to build an effective military organization (Collier
2000b; Lichbach 1995).
Both opportunity and motivation affect the length of conflict. The rebels need
motivation to keep on fighting. They also have to meet their financial needs. Group
cohesion is necessary to minimize deterrence and the risk that the group splits into
two or more competing rebel groups or becomes severely weakened by defection.
Conditions that facilitate cohesion may therefore provide another set of opportuni-
ties for rebellion.
Various models of conflict are related to this two-factor model of rebellion.
Collier and Hoeffler (1998, 2004) formulated their model around the concepts of
greed and grievance (see also Berdal and Malone 2000). In other writings, Collier
and Hoeffler stress the opportunity for rebellion. To some extent, they focus on
economic opportunity, which is akin to greed as a motive, but they also study the
geographical opportunity for rebels, notably rough terrain (forest cover and moun-
tains). Collier and Hoeffler recognize that both opportunity and motivation are nec-
essary for the outbreak of civil war. They argue, however, that a justice-seeking
rebellion has to overcome a set of challenges that a rebellion based on greed does
not have to address as long as it has the opportunity to organize a viable rebellion
(Collier 2000b, Collier and Hoeffler 2004, 587–588). They support this argument
with an empirical analysis where they do not find civil wars primarily in countries
where there are many grievances, but rather in countries where there are large op-
portunities for challenging the government militarily. An analysis of the opportuni-
5
ties and motivation for rebellion is also central in the World Bank project on civil
war (Collier et al. 2003).
Similarly, in the study of interstate war, Most and Starr (1989, 23) posit that
decisions to go to war require opportunity and willingness. Willingness refers to
“the choice (and process of choice) that is related to the selection of some behavior-
al option from a range of alternatives”, while opportunity is a shorthand term for
“the possibilities that are available within any environment”. In essence, this is the
same model that we use to study civil war.
The two conditions of civil war can also be applied to the government side:
The government’s motive can be a narrowly self-interested wish to remain in power
or a more altruistic desire to keep the nation together and protect the existing
structure of governance. The government’s opportunities are influenced by the
available instruments of repression (including preponderance if not a monopoly on
armed force) but also by the minimization of the geographical factors that create
opportunities for the rebels.
Both motivation and opportunity is required for armed conflict to occur. The
argument reviewed above implies an interactive effect of the two theoretical con-
cepts: To the extent that we can quantify the extent of motivation and of opportuni-
ties for organized violence for a group, the risk of armed conflict is an increasing
function of the product of these two quantities:
risk of civil war =f (M·O)
Identity Several authors (e.g. Gurr 1970) see identity as a third factor in rebellion on a par
with motivation and opportunity. Ellingsen (2000, 229) identifies the same three
factors but labels them frustration, opportunity, and identity.
Whether or not a group of individuals share a common identity affects both
their opportunities and motivations for rebellion or – in the case of individuals loyal
to the government – for successful anti-insurgent warfare. A common identity is
essential for group formation. Regardless of the motives of the rebels, they will not
act together unless they see themselves as being in the same boat. A perception of
injustice perpetrated against someone is most likely to be a motivation for an
armed conflict if the injustice affects an identifiable group rather than isolated indi-
viduals.
A common identity also affects opportunity, however: A rebel group that re-
cruits its members from a social group with a long history of interaction may more
easily solve the collective action and coordination challenges that the organization
of an effective army poses (Gates 2002; Collier 2000b). A government also needs a
certain amount of coherence to fight a civil war. Lacking this, it may collapse at the
first rebel challenge.
Risk of onset, incidence, duration, and severity An analysis of motivation and opportunity is fruitful for the study of onset (when
civil war breaks out), for incidence (the proportion of country-years that have an
6
ongoing civil war), and for duration (how long the violence lasts). The incidence of
conflict is a function of the risk of onset and of duration – incidence is high in
countries where the risk of onset is large, or when conflicts tend to be long. Inci-
dence is highest where conflicts are both very probable and likely to last for a long
time. Moreover, motivation and opportunity affect how severe a war becomes. To
the extent that actors have motivation and opportunity for initiating violent conflict,
they will also tend to have incentives to prolong and escalate the fighting.
There are differences, however. If only a narrow group has the opportunity
and motivation to initiate and sustain an armed insurgency, the conflict may be
long but not particularly severe. The Basque and Northern Ireland conflicts are ex-
amples of such enduring but relatively low-scale conflicts.
Motivations may also shift during conflict. Collier (2000a) argues that rebel
group leaders are forced to rely on private incentives even when the initial motiva-
tion is to address injustices. This may cause the insurgency to shift its motivation
from “justice-seeking” to “loot-seeking”, and the factors that help explaining a con-
flict’s initiation may be less relevant for explaining its continuation.
Democracy and the conditions of violence
Democratic governance in itself can be seen as a conflict management system
where different interests meet and are resolved peacefully. Democracy is related to
civil war through motivation as well as opportunity.
Motivation For broadly-based opposition groups that organize to change a government’ s policy
regarding redistribution or cultural issues, effective, inclusive and responsive dem-
ocratic institutions reduces the motivation for armed resistance. Legitimate de-
mands have a good chance of being met through such institutions. Democratic
countries are in general less repressive than non-democracies (Zanger 2000). The
absence of repression and the assurance of some political rights remove some of
the motivation for rebellion in and of itself (Muller and Weede 1990).
However, democracy cannot remove motivations for rebellion for all potential
rebel groups. The IRA rebelled against the UK government over the right to secede,
and in Chile a military coup in 1973 removed not only the elected government but
the whole democratic system. The majority may perceive the political demands of
narrow opposition groups as too extreme, such as a demand for secession or an
alternative form of political system. In many circumstances these demands are not
compatible with the democratic system present or the system may not be sufficient-
ly inclusive to handle the demand. Such groups may have a motivation for armed
resistance even in democracies.
However, the narrower the group, the smaller the chance that it will succeed
through a military strategy. A group that is narrow enough to be excluded from pol-
itics will therefore be deterred by bleak expectations. But very narrow groups that
primarily seek private gains will not be less motivated to use force in democracies.
7
Opportunity Democracy may also offer an opportunity for the rebels to organize an insurrection,
given the greater openness and more liberal practices of democratic regimes. These
liberal practices do not only allow individuals more space and time, but also open
up for organization of political interests unknown under autocratic regimes. This
openness is most often used in liberal ways, such as trade and commerce, but un-
der some circumstances it can be detrimental to civil peace.
By allowing political organizations, the potential for rebellion is always pre-
sent, but it is intensified if there is a change in the political regime that limits the
organizational freedom. In Algeria a very violent conflict was spawned from a failed
liberalization process. The aim of that process was to create a modern-style democ-
racy but in the face of a likely fundamentalist victory a crackdown severely limited
the political freedoms.
On the other hand, democratic governance severely limits the opportunity of
democratic leaders. Through mechanisms often referred to as checks and balances,
democracy is as much about limiting executive power as selecting the executive
officer. After losing an election, a democratic politician has a choice between adher-
ing to the result and opting for another chance at the next election or mounting a
rebellion against the electoral outcome. A consolidated democracy is often defined
as a system where the latter alternative always is the worst option.
Summing up how democracy affects civil war through motivation and oppor-
tunity, we assume that the motive can be modeled as (1–D), where the degree of
democracy is measured on a scale from 0 to 1. But democracy may also offer an
opportunity for the rebels to organize an insurrection, given the greater openness
and more liberal practices of democratic regimes. Thus, we may model the oppor-
tunity for rebellion simply as D on the same scale of democracy. Ignoring other mo-
tives as well as geographic and economic opportunity factors, our simple model of
the risk of civil war becomes
risk of civil war = f(D(1–D)),
which yields an inverted U-curve between democracy and the risk of civil war.
Identity The political norms that keep democracies stable can be seen as a form of identity –
a democratic identity. This form of identity is not formed overnight, but is shaped
through positive experiences of democratic governance. In contrast, the formal de-
mocratization of a country through the setup of electoral institutions can happen
much faster. This creates a potentially dangerous situation where other identities
than the democratic identity play important roles. Snyder (2000) has shown how
nationalism in fragmented societies can lead a process of democratization into civil
war, as it did in post-Tito Yugoslavia.
8
We assume that the identity of a coherent polity is stronger than an incoher-
ent one. However, we believe that a regime with a high level of democracy is more
coherent than one with a high level of autocracy. In the former, political challengers
will aim to take over the government, but not to change the political system. The
incumbent will yield to the response of the electorate and will step aside in an or-
derly transition. In autocracies, regime changes are rarely that orderly. The regime
is more likely to change along with the power-holders and violence is more likely.
To the extent that identity is seen as a separate factor, it should reinforce the in-
verted U-shaped relationship.
Hypotheses
Level of democracy These considerations regarding how democratic institutions affect motivation and
opportunity give rise to a set of hypotheses. The most general expectation is that
democracy reduces the motivation for conflict. At the same time, however, democ-
racy increases the opportunity of conflict. We argued above that the risk of civil war
is proportional to the product of motivation and opportunities for a country’s poten-
tial rebels. Democratization in a non-democracy has therefore two counteracting
effects, and the combination of motive and opportunity produces an inverted U-
shaped relationship between democracy and rebellion. In other words, in strictly
authoritarian states, the probability of rebellion is low, since the opportunity is
close to zero. In near-perfect democracies, the motive for rebellion is close to zero,
so the probability of rebellion is low. In the in-between area of semi-democracy (or
semi-autocracy), the probability of rebellion is the highest. Applying the same mod-
el to the government side, yields the same result. Coherent polities, whether auto-
cratic or democratic, have a strong motivation for maintaining the regime, whether
for egotistical or altruistic reasons. Incoherent (or inconsistent) polities in the mid-
dle have a much weaker sense of purpose. They are also likely to be weaker in both
normative and repressive power. Thus, the likely reaction of the government side
reinforces the inverted U-shape posited for the probability of rebellion.
Another implication of the two counteracting effects is that if a group is suffi-
ciently motivated and dedicated to start a rebellion against a democratic regime, we
should expect this conflict to be more durable than under other regimes. Not only
are groups that use armed force against democratic regimes particularly motivated,
but their democratic opponents are less capable adversaries as they are unable to
use the wide range of counterinsurgency methods available to autocrats, such as
mass killings and collective punishment. The expectation that democracies use rel-
atively mild counter-insurgency measures makes us expect that conflicts in democ-
racies will be less severe but also more durable.
H1: Democracy is negatively related to the onset and severity of civil war, but posi-tively related to duration. Democracy has an indeterminate effect on incidence.
9
H2: Democracy relates to the onset, incidence, duration, and severity of civil war through an inverted U-curve: semi-democracies are most prone to violence.
Change in institutions We also relate change in political institutions to civil war. Autocratic countries do
not become mature consolidated democracies overnight but usually go through a
rough transition. The transition process opens up opportunities for potential re-
bels. de Tocqueville, writing on the French revolution, (1856/1955, 182) points out
that “revolutions do not always come when things are going from bad to worse. . . .
Usually the most dangerous time for a bad government is when it attempts to re-
form itself”. The same observation holds true for the Russian revolution and subse-
quent civil war.
Even when the change is in the direction of democracy, it may not be suffi-
cient to reduce all relevant groups’ motivation for war. Autocrats are likely to re-
lease their grip on society only when they find the status quo to be unsustainable,
and will not reform more than they find necessary. Przeworski (1991) argues that a
partial liberalization will be received as too little in the eyes of civil society, which
will give the autocrats a choice between caving in to the demands present or to re-
treat into an excessively repressive regime. Democratization may also motivate for-
mer elites to instigate coups or other armed attempts at reinstating the former
status quo. If the direction of change is toward autocracy, the deconsolidation of
political institutions also implies increasing repression (Zanger 2000, 225–226). In
turn, such repression is likely to promote civil war (Lichbach 1987, 1995; Moore
1998).
Huntington (1991, 192ff.) also finds political violence to be coupled with de-
mocratization. Autocratic incumbents are unlikely to yield power without some re-
sistance, which could result in serious conflict. Communal groups in liberalizing
autocracies have substantial opportunities for mobilization, but such states usually
lack the institutional resources to reach the kinds of accommodation typical of es-
tablished democracy (Gurr 1993, 165). When authoritarianism collapses and is fol-
lowed by ineffectual efforts to establish democracy, the interim period of relative
anarchy is ripe for ethno-national or ideological leaders who want to organize rebel-
lion, as several post-communist states have experienced.
The initial high level of uncertainty and unrest caused by democratization will
gradually diminish as protesters abandon their aspirations or find ways to obtain
part of what they want within the new regime. In the case of democratization, new
and more open institutions take root and promote a peaceful resolution of domestic
conflict. As time passes, these become more entrenched, and the likelihood of re-
gime failure decreases. The pattern works similarly for autocratization. As repres-
sive institutions strengthen, the effect of the regime change is less destabilizing and
therefore less likely to generate political violence.
Semidemocracies are politically less stable than either autocracies or democ-
racies (Hegre et al. 2001, Table 1, 38; Gates et al. 2006). The hypothesis that low
political stability predicts to civil war is therefore consistent with an inverted U-
10
shaped relationship between civil war and the level of democracy. We hypothesize,
however, that a curvilinear effect of democracy on civil war persists even when con-
trolling for political instability.
H3: Political instability is positively related to the onset, incidence, duration, and se-verity of civil war, but most clearly for onset.
H4: The curvilinear effect of democracy on the onset, incidence, duration, and severi-ty of civil war persists even when controlling for political instability.
Elections In democracies, elections provide the main focal points for political change. The
difference between winning and losing an election is larger in some societies and
smaller in others. In consolidated democracies losing an election simply means that
one has to wait an election period and try again. In other societies, this option is
less certain, since an incumbent may cancel or rig a subsequent election. We posit
that losing an election can provide strong motivation for rebellion if the loss is ac-
companied with a high distrust of the winner. This leads us to predict that holding
all other political factors constant, violence should be more likely right after an
election – or just before, in a pre-emptive strike by a likely loser. However, as more
and more elections are held within a political regime, the less reasonable it seems
to suspect the likelihood of another election. We therefore posit a more precise ex-
pectation. When the government of a new regime faces an election, the uncertainty
regarding democratic procedures is likely to be the highest. We single out this first
contest as the point most likely of experiencing armed conflict.
H5: The risk of civil war is higher in conjunction with elections, and particularly so for the first contested election in a new political regime. Elections also tend to be accompanied by a lower risk of conflict termination and increases in severity.
Territorial vs. governmental conflict Several scholars have argued that different kinds of civil war can have different ex-
planations. A violent coup differs dramatically from a secessionist conflict. Fearon
and Laitin (2003), for example, perform separate analyses of ethnic civil conflicts.
Starting from the distinction between territorial and government conflicts in the
Uppsala/PRIO conflict data, Buhaug (2006) argues that the nature of the oppor-
tunity determines the direction of the conflict. Large countries are more likely to
experience conflicts in peripheral areas where rebels will fight for secession. Cap-
turing the central government is hard and may not even be necessary to redress
the rebels’ grievances. In such cases, opportunity and motivation are present for
territorial conflicts, but not for governmental ones.
Smaller countries are generally less suitable for separatist insurgency, but in
such countries, capturing the government is also a more realistic option. Ethnic
diversity and rough terrain also offer good grounds for secession, whereas institu-
tional consistency is particularly effective at preventing conflicts over state appa-
ratus. We therefore hypothesize that the relationships discussed above between the
11
level of democracy and onset, incidence, duration, and severity of civil war will be
more pronounced for governmental conflicts.
H6: The effect of democracy on the onset, incidence, duration, and severity of civil war is more pronounced for government conflicts than for territorial conflicts.
Efficiency of institutions Democratic institutions primarily reduce the risk of armed conflict by addressing
grievances – by allocating government funds to widely useful public goods, by en-
suring equitable redistribution, or by granting individuals freedom of choice in reli-
gious and cultural issues.
Democracies vary in terms of how successful they are in delivering these poli-
cy outcomes, however. A well-functioning parliament that produces optimal deci-
sions may not succeed in averting violence if decisions fail to be implemented by
inefficient and corrupt bureaucracy, or if the government has only a limited pres-
ence in the home region of potential insurgents. Some students of the democratic
peace at the interstate level have argued that the effect is likely to be stronger for
highly developed societies, while for low-development countries it will be weak or
even totally absent (Hegre 2000; Mousseau 2000; Mousseau, Hegre, and Oneal
2003). A similar argument can be made for civil war (Hegre 2003, 2005): The moti-
vation and opportunity for rebellion are not determined only by the value of a terri-
tory, or a group’s financial and organizational strength, but also of the usefulness
of armed force to obtain control over the valuables. The mobility of capital in highly
developed democracies may reduce the incentives for (territorial) conflict – it may be
profitable to use armed force to gain control over a diamond mine or a rich agricul-
tural province, but not to gain control over Silicon Valley. This affects potential re-
bel groups as much as governments. On the other hand, a highly developed non-
democracy is likely to experience conflicts over the political system itself – a well-
educated citizenry has both motivation and opportunity to fight for democratiza-
tion.
Moreover, an educated and informed citizenry is essential to making demo-
cratic institutions fully effective in constraining the political leadership, so that it
carries out the policies that the voters prefer. Good systems of education and the
free flow of information may in turn be dependent on economic development. In
addition, it is clear that high-income democracies are much more stable than low-
income democracies (Lipset 1959; Przeworski et al. 2000; Gates et al. 2006). Confi-
dence in the ability of democratic institutions to provide just and equitable distri-
bution of resources is likely to decrease “grievance”. It is clear that this confidence
should be larger the more stable the institutions are, and hence greater in high-
income democracies. In terms of our opportunity and motivation framework, the
citizens of highly developed democracies have greater opportunities for rebellion,
but this is offset by less powerful motives to do so. In addition, the opportunities
for the government to use armed force at the expense of own citizens are more lim-
ited in highly developed countries. We therefore hypothesize that the posited rela-
12
tionships between civil war and the level of democracy will be stronger the higher
the level of development.
H7: The effect of democracy on the onset, incidence, duration, and severity of civil war is stronger for developed countries.
Democratic neighborhood Most civil wars have transnational dimensions. Murdoch and Sandler (2002) and
others have found economic spillovers from civil wars to proximate countries, and
there are many instances where political instability in one country destabilizes
neighboring countries. For example, the 1997 conflict in DRC was influenced by
the civil war in Rwanda. One of the possible mechanisms behind the spatial diffu-
sion stems from the character of the political system in neighboring countries.
Democratic government is likely to have an effect on neighboring states by setting
an example and sometimes by coercion. Rebels in autocratic countries are likely to
be motivated by democratic rights won in neighboring countries. Democratic gov-
ernments may seen neighboring autocracies as a threat and may decide to support
rebel movements overtly or covertly in order to weaken or overturn the regime.
Gleditsch (2007, 298) hypothesizes that the less democratic the political institu-
tions of neighboring countries, the higher the risk that a country will experience a
civil war.
Here, we take a slightly different starting-point, inspired by research on the
interstate democratic peace, which indicates that mixed political dyads have the
highest risk of war (Gleditsch and Hegre 1997). Rather than attacking its neighbor
directly, a government can fuel a rebellion, and thereby fight through a proxy. The
Indonesian involvement in the Malayan civil war in the 1960s is a clear case. This
leads us to expect that civil war is less likely in democracies when neighboring
countries are also democracies, but more likely in autocracies and semi-
democracies.
H8: A democratic neighborhood is negatively related to the onset, incidence, duration, and severity of civil war in a democracy and positively related to civil war in a semidemocracy and an autocracy.
Inclusiveness of democratic institutions Democracy can take different forms. Reynal-Querol (2002) finds that more inclusive
forms of governance are less likely to experience violence. She defines inclusiveness
as the maximum distance between the preferences of any population group and the
aggregated preference produced by the political system. Her finding is based both
on comparisons of different electoral systems and on the presence of so-called veto
players – institutional constraints on the executive branch. The logic behind this
argument rests on Downs’ (1957) economic theory of conflict and asserts that while
a majoritarian system overall produces the most efficient policies, this efficiency
might damage a minority to the extent that it chooses to rebel against this policy. A
more inclusive system, such as those based on proportional representation will be
more representative and therefore build larger compromises, reducing the motiva-
tion for rebellion.
13
This view, however, is not unopposed. Not all political systems are best de-
scribed by the model presented by Downs. In a situation where a society is split
along a single issue, such as identity, ideology or religion, all political parties must
communicate a clear and unambiguous stance on this issue. A proportional repre-
sentation system will include more parties and most likely increase the level of
competition. Writing on Indian politics, Wilkinson (2004) shows how intense politi-
cal competition over issue salience drives these parties to violent means in order to
focus the political agenda on the ethnical cleavages. Extremist tactics polarize these
societies further. This might benefit the extreme parties, but for the political system
as a whole it results in a dangerous political outcome far from the median voter’s
preference.
While the two views differ with regard to which set of institutions should be
more likely to experience conflict in the first place, they agree that conflicts in more
inclusive systems should be shorter as well as less severe. In addition to testing the
effect of proportional representation, as does Reynal-Querol, we want to look at fed-
eralism and parliamentarism (as opposed to presidentialism). We assume that
these more inclusive forms of democracy are likely to have a greater dampening
effect on civil war. We hypothesize that this is true for the onset, incidence, dura-
tion, and severity of civil war.
H9: The effect of democracy on the onset, incidence, duration, and severity of civil war is more pronounced for inclusive types of democracy.
The impact of the Cold War Finally, we assume that the various relationships between democracy and civil war
are generally strengthened after the end of the Cold War. The reason for this is the
polarization during the Cold War, which led Western democratic powers to support
autocratic, non-communist regimes as the lesser evil. This external support re-
duced the opportunities for insurgency against otherwise weak governments. The
East and West supplied proxy wars in Angola, Central America, and elsewhere and
most of these conflicts ended when the outside support dried up. After the Cold
War, rebel groups have been forced to a much larger extent to rely on the opportu-
nities that are determined by domestic conditions, such as the availability of loot-
able resources (Collier and Hoeffler 2004) and weak governments (Hegre et al.
2001; Fearon and Laitin 2003).
Such polarization has occurred in other world-encompassing conflicts, too.
During the fight against Nazism, Western democracies tolerated (or even supported)
communist regimes and insurgencies as a lesser evil than Nazism. In the present
“war on terror”, autocratic regimes in the Middle East are more palatable than radi-
cal and anti-Western Islamic regimes. Yet, most of the time after 1989 has been
less polarized than the Cold War years. We might expect to find the world reverting
to greater polarization after 2001, but this period is too short to allow testing of
such a hypothesis.
14
H10: The relationships specified in the previous hypotheses are generally strength-ened after the end of the Cold War.
Previous empirical studies
A number of studies have addressed these issues. Collier and Hoeffler (2004) dis-
missed democracy as an influence on the onset of civil war, but their research de-
sign with five-year periods is not well suited to study the impact of political factors.
Fearon and Laitin (2003) and others also fail to find a linear relationship between
democracy and civil war. Carey (2007), however, found that countries with different
forms of executive election had (in varying degrees) lower risks of large-scale violent
dissent (using data from Banks 2000) than countries with no elections. Several
studies have also concluded that there is no relationship between democracy and
the duration of civil war (Collier, Hoeffler, and Söderbom 2004, De Rouen and So-
bek 2004; Fearon 2004). However, Elbadawi and Sambanis (2000) found democra-
cy to be negatively related to the incidence of civil war in Africa and Elbadawi and
Sambanis (2002) found the same for a global study.
The inverted U-curve between the level of democracy and civil war was re-
ported by Muller and Weede (1990) and Ellingsen and Gleditsch (1997). In a model
with an extensive set of control variables, Hegre et al. (2001) found the inverted U
for the onset of civil war for the entire Correlates of War period (1816–1992) as well
as for the post-World War II period (1946–92). Hegre et al. also found a minor and
non-significant negative linear effect of democracy on civil war and a positive and
significant effect of political instability, as measured by the proximity of the most
recent regime change. The inverted U-curve has also been found in global studies
by Sambanis (2001), de Soysa (2002), Reynal-Querol (2002), Bates et al. (2003),
Smith (2004), Urdal (2005), and Bussmann and Schneider (2007), but not in Hegre,
Gissinger & Gleditsch (2003). A similar finding is reported by Auvinen (1997).
Fearon and Laitin (2003) found that a dummy variable for anocracy (semi-
democracy) had a positive influence on civil war, consistent with the inverted U hy-
pothesis. Krause and Suzuki (2005) report an inverted U-curve for Asia as well as
for Sub-Saharan Africa, as do Henderson & Singer (2000) for the post-colonial
states of Africa, Asia, and the Middle East. Bussmann, Schneider & Wiesehomeier
(2005) report an inverted U-curve for Sub-Saharan Africa, but with such a high
turning-point that most of the countries are on the upward slope of the curve.
The inverted-U finding relies mostly on the coding of the Polity democracy in-
dex. Treier and Jackman (2005) argue that measurement errors in the index, espe-
cially at the two ends of the scale, necessitate the use of correction methods. This
problem is particularly acute when using a non-linear specification of the variable.
In a reanalysis of Hegre et al. (2001), they conclude that there is no significant rela-
tionship between democracy squared and civil war onset. However, unlike Treier
and Jackman we think that the measurement error is greatest in the middle rather
than at the end points of the scale. Pure and consistent regimes are easily identi-
fied, while the exact nature of a mixed regime is hard to assess. Thus, we question
whether measurement error biases the result in the way that they claim.
15
A more serious problem with the Polity index is pointed out by Hegre et al.
(2001), Vreeland (2008), and Strand (2006: Ch. 5). If there is “factionalism” in a
country with democratic institutions, e.g. intense inter-group conflicts that may or
may not be violent, the Polity index will code the country as an imperfect democra-
cy. Hence, the finding that imperfect democracies have more civil war may have a
tautological element. Vreeland and Strand also reanalyze Hegre et al. (2001), and
find the evidence for the inverted U to be much weaker when removing the two
components of Polity that contain this coding. In an extensive analysis, Hegre and
Sambanis (2006) also find robust evidence for the inverted U-shaped relationship
when using the Polity variables. They do not find any relationship when replacing
these with variables that remove the “factionalism” component. We return to this
issue in the empirical analysis.
There are few studies of the relationship between elections and subsequent
violence. Bates et al. (2003) find that in particular the second election is a critical
turning point in partial democracies. Strand (2006: Ch. 8) corroborates this finding,
and finds a conditional relationship between proximity to an election and a height-
ened risk of conflict onset.
Hegre (2003) found strong evidence that democracy is correlated with civil
peace only for developed countries and for countries with high levels of literacy.
Conversely, he found that the risk of civil war decreases with development only for
democratic countries. Buhaug (2006) introduced the distinction between civil wars
over territory and government and found the inverted U-curve for the latter but not
for the former. Gleditsch (2002a, 106; 2007) found that the degree of democracy in
neighboring countries had a significant effect on the risk of civil war, in fact greater
than the country-specific effect of democracy.
As noted, Reynal-Querol (2002) found empirical support for the conflict-
dampening effect of more inclusive types of democracy, as did Schneider and
Wiesehomeyer (2005). Carey (2007) found that countries with executives elected
through multiparty elections have lower risk of insurgency than countries without
elections. But so do countries with single-candidate elections. Hartzell and Hoddie
(2003) found that power sharing among former combatants specified in a peace
agreement increased the likelihood that peace will endure after a civil war. They
suggest that this occurs because of the unique capacity of power-sharing institu-
tions to foster a sense of security among former enemies and encourage conditions
conducive to a self-enforcing peace. Binningsbø (2006) also found that power shar-
ing – and particularly the formation of a grand coalition – increased the duration of
peace after the end of a conflict. However, Hegre and Sambanis (2006: 526) found
that presidentialism was associated with a lower risk of civil war onset.
Using a new dataset on battle deaths (Lacina and Gleditsch 2005), Lacina
(2006) found the determinants of conflict severity to be quite different from those
for conflict onset. Democracy, rather than economic development or state military
strength, is most strongly correlated with fewer deaths.
16
Data
Since the content of democracy itself is a highly debatable subject, it is not surpris-
ing that a large number of operationalizations are found in the academic literature.
Despite some criticism (e.g. Gleditsch and Ward 1997), the empirical literature in
international relations and peace research has overwhelmingly chosen to use the
data from the Polity project (Jaggers and Gurr 1995; Marshall and Jaggers 2003).5
The empirical confirmation of the inverted U is largely based on Polity. We start our
analysis using this dataset, but most of our analyses use the SIP index from the
MIRPS6 dataset (Gates et al. 2006), which differs from Polity in that the participa-
tion dimension is based on reported election turnout as reported in Vanhanen
(2000).
The civil war data all come from the Uppsala/PRIO conflict dataset. We rely
on onset data from Strand (2006: Ch. 4); incidence data from Gleditsch et al.
(2002), most recently updated in Harbom, Högbladh, and Wallensteen (2006); dura-
tion data from Gates and Strand (2006); and data on severity (battle-deaths) from
Lacina and Gleditsch (2005). We include all the conflicts in the Uppsala/PRIO da-
taset, which has a lower level for inclusion at 25 battle-deaths in a given year.7
Proximity to regime change is based on the SIP measure, where a change is
defined as either a halving or doubling of participation; a change in the institution-
al recruitment of the executive (i.e. whether the executive is elected or not); or a
change in Polity’s executive constraints dimension of more than one unit. Political
difference is defined as difference between the democracy level as measured by the
SIP index of a given country and the average of all non-transitional regime values
within a distance of 100 km that meet the Gleditsch and Ward (1999) system
membership criterion. The variable is continuous and ranges from –1 (neighbors
are democratic while the country itself is non-democratic) to 1 (the country is dem-
5 Many economists prefer to use the data on political rights from Freedom House (annual) and a few scholars use the Polyarchy data developed by Vanhanen (2000). For a survey of nine measures of democracy, see Munck and Verkuilen (2002). 6 SIP = Scalar Index of Polities, MIRPS = Multidimensional Institutional Representation of Political Systems (Gates et al. 2006). 7 Most of the analyses have also been tested on the more restrictive measure of 1,000 or more battle-deaths in a year (which is defined as ‘war’ in this dataset as well as in the Correlates of War data). Generally, the results are very similar, so we do not comment in detail on the alternative analyses.
17
ocratic while the neighbors are non-democratic. We include this variable and its
square in the analyses below. The square term ranges from 0 to 1 and reflects the
difference between the level of democracy in the country and among its neighbors.
Data on election dates are from (Strand 2006: Ch. 8). Our expectation, in
line with the finding of Bates et al. (2003), is that violence is most likely in connec-
tion with the first election in a new political regime where the incumbent govern-
ment is at risk of losing office. We will therefore include two election measures. One
variable measures the proximity to the election that provides the first test of the
new regime, and the other variable measures proximity to any other election. Data
on presidentialism, proportional representation, and federalism are scored accord-
ing to Schneider and Wiesehomeier (2005), partially based on data from Golder
(2005). For economic development and population we use data from Penn World
Tables, but in order to reduce missing data problems, we use the expanded data
from Gleditsch (2002b). Both of these indicators are lagged one year and log-
transformed. The data on ethno-linguistic fractionalization were collected by Roeder
(2001). For the oil dependency variable, we use data from Fearon and Laitin (2003).
Analysis
We start by a bivariate analysis of the relationship between the various democracy
variables and four measures of armed conflict: onset, incidence, duration, and se-
verity (Table I). All figures are in ratio form and roughly comparable. These ratios
tell us how many times better or worse democracies perform in comparison to other
regimes. The column labeled “onset” provides estimates of the relative risk of onset
of armed conflict – the estimated probability of onset of conflict when the explana-
tory variable is at a value X1=X0+1 divided by the estimated probability when the
variable is at X0. For variables that range from 0 to 1, this is the same as the risk
when the variable is at its maximum divided by when it is at its minimum. The col-
umn labeled “incidence” give estimated ratios of the odds of a country being in-
volved in a conflict in a given year. The column labeled “duration” gives an estimate
of relative durability of a conflict given that it has started. The column labeled “se-
verity” reports the estimated ratio of the number of annual battle deaths during the
18
war when the explanatory variable is at its maximum divided by the number of
deaths when it is at its minimum.8 For the incidence, duration, and severity anal-
yses, variables are measured independently for every ongoing year of conflict.
_____________________
Table I in about here
_____________________
The bivariate results are very favorable regarding the effects of democratic
governance. The first line shows the ratios for countries with maximum Polity IV
score as compared to countries with minimum Polity score. The .61 estimate in the
“onset” column shows that democracies are 39% less likely to experience a civil war
onset than autocracies. The .95 estimate in the “incidence” column reflects that
they are 5% less likely to be in a state of civil war at a given point in time. However,
once a conflict has erupted in a democracy, it tends to persist. In fact, democracies
experience conflicts that are almost four times longer than do autocracies. On the
other hand, the “severity” column shows that conflicts in democracies are less in-
tense – in each year of the conflict, they claim only a third of the number of victims
as similar conflicts in non-democracies. Overall, armed conflicts in democracies
would seem to claim about as many lives as conflicts in non-democracies, but they
are spread out over a much longer period. This observation does not imply that as
many people have been killed in democracies as in non-democracies. Keeping in
mind that armed conflicts are much less common in democracies than other re-
gimes, we can still conclude that inhabitants in democracies are generally safer.
Indeed when we look at the period 1946–2002, we find that fatalities in non-
8 The relative risk of onset estimates were obtained using bivariate Cox regression with calendar time as the time variable (see Raknerud and Hegre 1997 and Hegre et al. 2001 for a complete description). The odds of incidence ratios where estimated using logistic regression on a country-year dataset. The estimated relative risk of conflict termination was obtained using Weibull regression on a conflict du-ration dataset, using duration time as time variable. The estimated battle deaths ratios were obtained by estimating an OLS model with log battle deaths as the dependent variable, and taking the anti-log of the estimated coefficient. Full documentation of the design of the statistical tests is given in an appendix that will be released with the replication data (Strand 2007).
19
democracies exceed those in democracies by three to one. All ratios are significantly
different from 1.
The second line in Table I shows similar results for the square of the Polity IV
index. The index was squared and rescaled such that the squared index ranges
from 0 (for the midpoint) to 1 (for the endpoints of the Polity scale). The ratios pre-
sented therefore compare the endpoints of the Polity scale to the mean. We see that
the inverted U-curve is strong and robust in a bivariate model, both for “onset” and
for “incidence”: Consistent democracies and autocracies are 78% less likely to have
conflict onset and 70% less likely to be at war at any given time. Consistent autoc-
racies and democracies also have longer wars than the inconsistent regime types,
but this result is not statistically significant. Finally, inconsistent regimes have
three times more lethal wars than pure autocracies and democracies.
Lines 3 and 4 present the same set of results using the SIP democracy index
from the MIRPS project. The SIP index avoids the endogeneity problem in Polity by
replacing the participation component of Polity with the corresponding element of
Vanhanen’s Polyarchy index. Hence, this index is not affected by the “factionalism”
problem in Polity. The results for the SIP index (line 3) are roughly the same as for
the Polity index. The results for the SIP democracy index squared, however (line 4)
are clearly weaker than the corresponding for the Polity index. The exception is the
duration analysis, where we find inconsistent regimes to have almost five times
longer wars than consistent ones. Overall, in a bivariate analysis, there seems to be
significant support for an inverted U-shaped relationship.
Both Polity and MIRPS combine information on how political power is gath-
ered and how it can be used. Vanhanen’s Polyarchy dataset focuses exclusively on
how power is won, through observing election outcomes. The Polyarchy Index of
Democracy is the product of electoral participation and competition. Using this in-
dicator of democracy, we find a very strong support for both a linear and curviline-
ar relationship between democracy and onset, incidence and severity, but weaker
results for duration. When we use the Freedom House indicator, whose particular
strength is to measure how political power is used, we find fairly similar results.
The one clear difference is that there is no curvilinear relationship between democ-
20
racy and severity. The Freedom House scores are partly based on freedom from
“acts of violence or terror due to civil conflict or war”, which makes this dataset,
too, biased in favor of our hypothesis (Freedom House, 2006).
Table I shows that the bivariate relationship between level of democracy and
onset of conflict is very robust across different definitions of democracy: Democra-
cies experience fewer conflict onsets than other regimes. Once conflicts occur, they
are less brutal in democracies than in other regimes. Figure 1 shows the bivariate
relationship between regime type and civil war, with the size of the circles indicat-
ing the severity of the conflict, plotted on a background of a tripartite division of
regime type. Many of these conflicts have gone on for a long time but most of them
have claimed relatively few lives9. Using Polity and MIRPS, we find that autocracies
experience the longest conflicts, but these findings are not supported by Polyarchy
or Freedom House. There are aspects of the latter two datasets that weakens our
confidence in them. First, Vanhanen’s dataset does not take into consideration how
much power the executive has once it is won. This limits its ability to discriminate
between liberal and illiberal democracies, and it is this distinction in particular
that, according to our theory, makes democracies less capable of fighting rebel-
lions. The Freedom House dataset is only available from 1973, while the other da-
tasets span the whole period 1950–2000. This creates a particular problem for
duration analysis, since the most durable conflicts, which started early in the peri-
od, are excluded. Thus, the conflicts analyzed with the Freedom House data are a
sample of shorter conflicts and therefore not representative. These problems, and
the fact that regime changes are not dated in these datasets10, make Polyarchy and
Freedom House less suited for our purpose. We will not use these indicators in our
multivariate analyses.
Proximity to regime change is measured using a decay function. This function
assumes that the negative impact of a regime change is at its peak immediately
after the change takes place, after which it diminishes at a constant rate over time.
The decay function is therefore always 1 when time since last regime change is 0
9 Major exceptions being Sri Lanka and Turkey. For details, see the Appendix. 10 See the unpublished appendix posted with the replication data (Strand 2007) for an explanation of why this is problematic.
21
(i.e. the day before), and the decay takes the value 0 when the time since last re-
gime change is infinitely long. The attenuation rate is determined by the half-life
parameter, which in this analysis we have set to 4 years. Thus, four years after a
regime change, the risk of conflict is 50% of the original risk stemming from the
regime change. After 8 years, it is 25%, after 12 years 12.5%, etc. Whereas regime
change does seem unrelated to the chances that a conflict ends, it is clearly related
to the risk of its onset. Regimes that recently have experienced a regime change are
2.5 times more likely to see a conflict than regimes that have remained unchanged
for several years. Regime changes also lead to a doubling in the violence levels. The
incidence of conflict is also increased, but given the insignificant effect on duration
this effect seemingly stems from the increased risk of experiencing a new conflict
rather than the prolongation of present conflicts.
The nature of the political systems in the immediate neighborhood is clearly
also important. We do not find a particular high risk of war onset in countries that
have political systems that are different from the systems in the immediate neigh-
borhood.11 Just as we find an effect of level of democracy on duration, however, we
find that if a country is considerably more democratic than its neighborhood the
duration of armed conflict increases by a factor of 4. We also find that the incidence
of conflict is 29% higher for countries in heterogeneous neighborhoods than for
those in homogenous ones. On the other hand, wars are less intense than wars in
homogenous neighborhoods if the country is more democratic than its neighbors.
Countries that are politically different from their neighborhood also have less lethal
wars. This result cannot be explained by the level of democracy in the country it-
self, however, since the square of the political difference is also significantly less
than 1. Wars that take place in a country with a political system different from its
neighborhood lead to only one sixth of the annual fatalities compared to wars in
politically similar neighborhoods. This applies regardless of whether the country in
question is very democratic or autocratic.
11 In an analysis not reported in detail here, we find that a country – irrespective of its domestic sys-tem – in an all-democratic neighborhood has a 62% lower risk of conflict onset than a country in an all-autocratic environment. Indeed, the effect of living in a democratic neighborhood seems stronger than the effect of being a democratic country.
22
Among democracies, parliamentary systems experience more severe wars
than both presidential and mixed systems. Proportional representation systems
have a lower risk of war onset and may be less lethal, but there are indications that
they last longer than majoritarian systems. Finally, countries with federal systems
have a higher risk of war onset than centralized systems.
Our earlier research has shown that inconsistent regimes are considerably
less stable than consistent regimes. Therefore, we need to assess whether or not
the inverted U holds up when we control for proximity of regime change. In Table II,
the first line repeats the bivariate results for conflict onset from Table I. The second
line shows the bivariate relative risk when controlling for “proximity of regime
change”. Controlling for political stability does not eradicate the effect of democracy
on conflict onset, contrary to the results in Vreeland (2008), but the effect for the
inverted U is somewhat weaker. A politically different neighborhood now increases
the risk of civil war onset. Since democratization is contagious (Gleditsch and
Ward, 2006), this effect may imply that the most stubborn political regimes are
toppled violently. An example of this is the transition to democracy in Rumania in
1989, which occurred after the transitions in the neighboring countries and be-
came quite violent.
_____________________
Table II in about here
_____________________
The next two lines in Table II investigate the hypothesis in Hegre (2003) that
the effect of democracy is dependent on the level of development by estimating the
relative risk of onset of conflict separately for high-income and low-income coun-
tries (defined as having an average income of less than $5,500 per year in real 1995
US dollars). Line 3 shows that the relationship between democracy and democracy
squared is very strong for developed countries. The relationship is much weaker
and not significant in low-income countries. The neighborhood and election varia-
bles have strong but insignificant coefficients in the high-development sample,
which means that the trivariate model is not very precise. The coefficients are
smaller for the low-development sample, where there is some evidence of a neigh-
borhood effect but no effect from proximity to elections.
Buhaug (2006) and Buhaug and Rød (2006) report that conflicts over gov-
ernment have different causes than territorial conflicts. In lines 5 and 6, we repro-
duce their finding by estimating the bivariate relationship separately for the two
types of conflicts. Whereas the danger of governmental conflicts clearly depends on
the level of democracy of the regime, there are no connections between the two de-
mocracy indicators and territorial conflict. The difference in the political neighbor-
hood, on the other hand, strongly increases the likelihood of a territorial conflict,
but not governmental conflict.
23
Finally, the argument that the absence of democratic institutions has become
a more important predictor of conflict after the Cold War, receives support in lines
7 and 8. For the post-Cold War period, the bivariate relationships between our de-
mocracy variables and conflict are very strong. Moreover, the difference to the polit-
ical system in the neighborhood increases the risk of war onset strongly after the
Cold War. The results for the Cold War period are weaker and not as significant,
except for proximity to elections. Here, we are probably capturing an effect of politi-
cal instability in former colonies.
In the analysis reported in Table III, we add several control variables that
might both explain levels and stability of democracy. In the table, we only report
the estimated coefficients for the democracy variables – these should be interpreted
as risk or severity ratios contingent on the control variables. (Tables IV and V report
the full set of results for some of the onset and severity models.)
_____________________
Table III in about here
_____________________
The results in Table I indicated that there are strong bivariate effects between
democracy and armed conflict. The contingent relationships in Table III are gener-
ally considerably weaker. In fact, only the squared democracy variables are signifi-
cantly associated with the risk of civil war onset, and none of the indicators are
related to duration. From this we can claim weak support for the hypothesis that
semidemocracies have a higher risk of conflict and that regional differences also
play a role in increasing this risk. The contingent relationships with war severity,
on the other hand, are strong and generally in the predicted direction. Both prox-
imity variables (regime change and election) are reported to have some effect on the
risk of conflict, but these findings are not significant. Both of these factors are re-
lated to political instability, and it is premature to use the lack of significance as an
argument against their importance. We will return to this point later.
As is evident in Table IV below, low economic development is a strong predic-
tor of armed conflict. We also know from other studies that economic development
is strongly associated with stable democratic governance (e.g., Gates et al. 2006).
Highly developed countries tend to be democracies and peaceful. This explains why
there is no relationship between democracy and onset or incidence in Table III: The
peaceful quality of democracies may be due to their status as highly developed ra-
ther than as democracies. We saw in Table II that the effect of political regimes is
much stronger in developed countries. Hegre (2003) finds this to be true, even
when controlling for other variables.
The results in Table III give us two different versions of the inverted U-curve
between conflict and democracy. Using the Polity IV dataset, we find evidence of an
inverted U-curve regarding onset of conflict. The contingent relative risk of conflict
in very democratic or very autocratic countries is .42 or 58% lower than in a com-
24
parable semi-democracy. As noted, this relationship is partly explained by the cod-
ing of “factionalism” in Polity. The same effect, using the SIP dataset, is 42%. The
difference between these two findings is due to the endogenous component of the
Polity dataset. We find no evidence supporting the hypothesis that democracies are
less at risk than other regimes, nor does the analysis support a claim that their
wars are longer.
For both democracy measures, the results are weaker when we look at inci-
dence than at onset, i.e. the coefficient for the squared term for democracy is closer
to one. As noted, incidence includes aspects of both onset and duration. Semi-
democracies experience fewer conflicts but their conflicts are more durable when
they occur. These two effects pull in opposite directions, and the net effect is inde-
terminate.
We saw in Table I that democracies are associated with less intense conflicts,
confirming the results in Lacina (2006). In Table III, the annual level of battle-
related fatalities is 45–60% lower per year for conflicts in democracies than in au-
tocracies, even when we take into account population size. In the multivariate
analysis, there is only weak evidence that democracies or inconsistent regimes have
longer wars than non-democracies. The analysis clearly shows that the average
democratic conflict is less violent than the average conflict in a non-democratic
country.
The effect of the political difference with the neighborhood is greatly reduced
when we take into account the control variables, including the level of democracy
in the country in question. The “political difference squared” is estimated to in-
crease the incidence of conflict but, as noted in the bivariate analysis, it tends to
strongly reduce its severity.
Comparing different democratic regimes, we find no real differences between
onset ratios, but the duration and severity analyses seem to echo the initial com-
parison of democracies and autocracies. Presidential regimes have more violent
conflicts. In countries with proportional representation, the conflicts tend to last
longer, but they are less violent per year than conflicts in majoritarian systems.
Since the conflicts are longer, the incidence of conflict is higher in parliamentary
systems. The findings for federalism disappear when we add control variables
(country size is particularly important here). Wars in federal democracies seem to
be shorter than in centralized systems, and this tendency is strong enough to be
reflected in a significant estimate in the incidence analysis.
Summarizing the results from Table III, we conclude that most of the bivari-
ate effects reported in Table I do not hold up when introducing control variables.
However, given that a conflict has erupted, there seems to be strong and persistent
effect from political regimes on the severity of the conflict. Democracies tend to ex-
perience more protracted conflicts, but this finding is not significant. On the other
hand, democracies have clearly less violent wars. We discuss this finding in more
detail below.
25
Several studies of the causes of conflict (Fearon and Laitin 2003, Hegre et al.
2001) report an effect from political instability, which seems to contradict the find-
ings reported in Table III. Table IV presents a more nuanced result, indicating
strong support for the hypothesis that political instability is related to conflict but
agnostic to the correct operationalization of this concept. The figures reported are
the results from a full, multivariate analysis of onset of armed conflict.
_____________________
Table IV in about here
_____________________
The figures in Table III are based on Model 1, where we see that the inverted
U-curve receives a fair measure of support while neither regime change nor election
proximity is significant. However, a third measure of political instability, proximity
to independence, is strong and significant. Proximity to regime change and proximi-
ty to independence share the same causal logic. It is the unconsolidated nature of
the political regime that is the key factor explaining the onset of conflict. A final
proxy for political instability is the institutional difference from the neighborhood.
Gates, et al. (2006) provide evidence for a strong destabilizing effect from such po-
litical isolation. As argued earlier, this can be a consequence of the spatial cluster-
ing of regime changes. As a proxy for political instability, it differs somewhat from
the other measures, as it captures the potential for regime change as much as
change that has already taken place. Squared difference from political neighbor-
hood has a quite strong and robust effect on the relative risk of conflict.
In Model 2 we remove the competing operationalizations of political instabil-
ity, resulting in a stronger and significant effect from proximity to regime change.
This effect provides good support for our hypothesis of a relationship between polit-
ical instability and conflict, and it is supportive of the argument made above. Politi-
cal instability has an effect, but it is unclear through which variables this
manifests itself.
To investigate this claim further, we use the robustness checks introduced in
Strand (2006: Ch. 4). In the analysis reported so far, we have coded a new onset
whenever the conflict was inactive for more than two whole calendar years. Inevita-
bly, this includes a number of onsets that are commonly interpreted as a continua-
tion and direct consequence of the previous conflict period. In Model 3, we re-
analyze our data with a stricter requirement: new onsets are recorded only after
eights years of inactivity. In Model 4, we include only those conflicts that exceed a
total 1,000 battle-related deaths.
These robustness checks provide clear support for the instability hypothesis.
With the stricter onset requirement (Model 3), we find that proximity to both inde-
pendence and succeeding regime changes are significant contributors to risk of
conflict. The inverted U-curve is present as well. However, neither elections nor dif-
ference from neighborhood are significantly associated with increase of risk in this
26
model, although the coefficient for difference from neighborhood is almost as strong
as in Model 1.
When we focus on conflicts that exceed 1,000 battle-related deaths (Model 4),
proximity to regime change has almost exactly the same effect as in Model 3, but
the uncertainty increases to just above the limit of significance. The substantive
effect remains the same. On the other hand, proximity to independence is reduced
from being very influential to an insignificant factor. Proximity to election is a very
potent predictor in Model 4, in contrast to the other models. As expected, it is the
first election in a political regime under the initial government that significantly
increases the risk of conflict. Since this election very often comes in the initial peri-
od of a new political regime, this variable and proximity to regime change can be
seen as measuring almost the same thing. The election variable is more sensitive to
time and the effect of an election is quickly reduced, whereas the proximity to re-
gime change variable is much more persistent. Both are decay functions, but have
half-life values of 6 months and 2.9 years respectively. However, they are both
measures of political stability, and Model 4 further strengthens our hypothesis.
Table V shows correspondingly detailed results for models with severity as
the dependent variable. The unit of analysis is a conflict-year, and the dependent
variable is the natural logarithm of the estimated number of battle deaths in that
conflict-year. We include the lagged dependent variable in the model. If the year
before the year of observation was a peace year, we set ln(battle deaths) to 0. The
interpretation of the estimates therefore indicates the change in battle deaths from
the previous year. The estimates are exponentiated to be comparable to the risk
and odds ratios presented earlier. The estimate for the SIP democracy index signi-
fies that given last year’s severity and the values for the other control variables, a
war in a democracy on average has 55% of the fatalities in an average autocracy.
_____________________
Table V in about here
_____________________
In Model 1 we use Polity, while in Model 2 we use the SIP index. We find a
strong conflict-dampening effect of democracy for both measures. Since the main
term is smaller than 1 in both models, there is no inverted U-relationship here. The
estimated relationship is stronger for the Polity index than for the SIP index, how-
ever.
The effects of the control variables are remarkably similar to those found in
the onset analysis. The same factors that increase the risk of a war onset, contrib-
ute to making the war more severe. A regime change during a war increase the ex-
pected number of fatalities with 33–63% in the first year. As before, this effect is
assumed to decrease thereafter at a constant rate with a half-life of 1 year. Wars
that occur in politically different neighborhoods are less lethal than wars in similar
27
neighborhoods, but this does not apply if the difference is positive, i.e. for democra-
cies in non-democratic neighborhoods. Ethno-linguistic fractionalization reduces
the severity of wars, possibly because coordination problems in both armies reduce
their efficiency and scope. Wars in high-income countries are less intense than in
low-income countries, and oil exporters have more violent wars than countries with
comparable incomes derived from other sources. The population variable is not sig-
nificant. This difference from the onset analysis is probably because population has
a smaller variance in the war sample than in the full sample, and that much of the
population effect is taken up by the lagged dependent variable.
Conclusion
The overall question asked in this chapter is whether democracies experience less
conflict than other regimes. “Less conflict” may refer not only to fewer conflicts, but
also shorter and less violent conflicts. We formulated ten hypotheses to test the
relationship between democracy and civil war in detail. In general, we obtain well-
defined results for our analysis of the severity of war, and somewhat less conclusive
ones for the study of the risk of onset. As in earlier statistical studies of civil war,
our results for the duration of civil war tend to be the least conclusive. When the
onset and duration analyses indicate that variables predict both frequent and long
civil wars corresponding analyses also show a high incidence incidence of war.
Democracies experience fewer conflicts than other regimes (Hypothesis 1),
but this effect seems to be caused by the fact that democracies also tend to enjoy
other pacifying qualities. In particular, most democracies are high-income coun-
tries with stable institutions. Controlling for other factors, we find no linear rela-
tionship between democracy and the risk of conflict onset. We find some support
for a curvilinear effect indicating that semi-democracies are more at risk of experi-
encing conflict, but it is not entirely robust to our choice of democracy indicator
(Hypothesis 2). When democracies have wars, however, they are clearly less violent
than wars in comparable non-democracies. This relationship appears to be mono-
tonic – inconsistent democracies may have a higher risk of civil wars, but they are
on average not as violent as those in non-democracies.
We find solid evidence for a general relationship between political instability
and conflict, but we find no single, robust operationalization of political instability.
We do not find evidence supporting the hypothesis that the inverted U-curve rela-
tionship is exclusively due to the inherent instability of semi-democratic regimes
(Hypotheses 3 and 4). On the contrary, the inverted U-curve is robust when we
control for political instability. However, when we control for other factors that are
also related to political instability, the inverted U-curve is somewhat more fragile
than the effect of instability in onset studies.
We do not find support for a bivariate relationship between elections and civil
war but when controlling for other factors, the first “real” election in a new political
regime appear to have a partial effect (Hypothesis 5). The effect of proximity to elec-
tions seems to be part of the political instability-conflict nexus.
28
We find evidence that the effect of democracy is much stronger when we look
at governmental conflicts compared with secessionist conflicts (Hypothesis 6), and
that the bivariate pacifying effect of democracy does not hold for low-development
countries (Hypothesis 7). We also study in detail the importance of the political in-
stitutions in the neighborhood. When controlling for other factors, there does not
seem to be a robust association between neighborhood and the onset, incidence,
and duration of wars. We do find clear evidence that civil wars in non-democracies
that are located in democratic neighborhoods are considerably less violent than in
other non-democracies. This is due to the level of democracy among the neighbors
rather than the political difference in itself, since we do not find a corresponding
effect of difference for democracies located in non-democratic neighborhoods (Hy-
pothesis 8).
We have also explored differences between different democratic institutions.
Again, the results are much more conclusive for the severity analysis than for the
other aspects of war. As expected, presidential systems seem more violent than
parliamentary ones, and majoritarian systems more violent than those with propor-
tional representation – but the latter may have longer wars.
Finally, we find that the effect of democracy has become much more im-
portant after the Cold War (Hypothesis 10).
Overall, we confirm that democracy is strongly related to various aspects of
civil war. Both level and stability of democracy contribute to the risk of conflict, but
once conflict has erupted, political institutions do not explain the duration of con-
flict. We find strong effects from democracy on the severity of conflict. Our results,
except for severity, are much weaker when we include control variables. Factors
that are known to contribute to democracy and democratic stability also robustly
contribute to peace, indicating that part of the relationship is spurious. Better the-
ory is needed to sort out the multivariate relationships and the role played by the
system of governance. Future research should explore how democracy, democratic
stability, and peace interact and strengthen each other.
References
Auvinen, Juha. 1997. “Political Conflict in Less Developed Countries, 1981–89.” Journal of Peace Research 34(2): 177–195.
Bates, Robert H., David L. Epstein, Jack A. Goldstone, Ted Robert Gurr, Barbara Harff, Colin H. Kahl, Kristen Knight, Marc A. Levy, Michael Lustik, Monty G. Marshall, Thomas M. Parris, Jay Ulfelder, and Mark R. Woodward. 2003. Political Instability Task Force Report: Phase IV Findings. McLean, VA: Science Applications Interna-tional Corporation.
Berdal, Mats, and David M. Malone (eds.). 2000. Greed and Grievance: Economic Agendas in Civil Wars. Boulder, CO & London: Lynne Rienner.
Binningsbø, Helga Malmin. 2006, “Power-Sharing and Postconflict Peace Periods”, paper presented at the 47th Annual Convention of the International Studies Association, San Diego, CA, 22–25 March, http://64.112.226.77/one/isa/isa06/.
Buhaug, Halvard. 2006. “Relative Capability and Rebel Objective in Civil War.” Journal of Peace Research 43(6): 691–708.
Buhaug, Halvard, and Jan Ketil Rød. 2006. “Local Determinants of African Civil Wars, 1970–2001.” Political Geography 25(3): 315–335.
29
Bussmann, Margit, and Gerald Schneider. 2007. “When Globalization Discontent Turns Violent: Foreign Economic Liberalization and Internal War.” International Studies Quarterly 51(1): 79–97.
Bussmann, Margit, Gerald Schneider, and Nina Wiesehomeier. 2005. “Foreign Economic Liberalization and Peace: The Case of Sub-saharan Africa.” European Journal of In-ternational Relations 11(4): 551–579.
Carey, Sabine C. 2007. “Rebellion in Africa: Disaggregating the effect of political regimes.” Journal of Peace Research 44(1): 47–64.
Collier, Paul. 2000a. “Rebellion As a Quasi-criminal Activity.” Journal of Conflict Resolution 44(6): 839–853.
Collier, Paul. 2000b. “Doing Well Out of War: An Economic Perspective.” in Mats Berdal, and David M. Malone (eds.), Greed & Grievance: Economic Agendas in Civil Wars. Boulder, CO: Lynne Rienner (91–111).
Collier, Paul, and Anke Hoeffler. 1998. “On the Economic Causes of Civil War.” Oxford Eco-nomic Papers 50(4): 563–573.
Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers 56(4): 563–595.
Collier, Paul, Lani Elliot, Håvard Hegre, Anke Hoeffler, Marta Reynal-Querol, and Nicholas Sambanis. 2003. Breaking the Conflict Trap. Civil War and Development Policy. Ox-ford: Oxford University Press & Washington, DC: World Bank,
http://econ.worldbank.org/prr/CivilWarPRR/. Collier, Paul, Anke Hoeffler, and Måns Söderbom. 2004. “On the Duration of Civil War.”
Journal of Peace Research 41(3):253–274. Cunningham, David, Kristian Skrede Gleditsch, and Idean Salehyan. 2007. “It Takes Two: A
Dyadic Analysis of Civil War Duration and Outcome”, revised version of paper pre-sented at the GROW-net Workshop, Oslo, 10–11 February 2006, http://privatewww.essex.ac.uk/~ksg/papers/CGS_dyadic.pdf.
DeRouen, Karl R., and David Sobek. 2004. “The Dynamics of Civil War Duration and Out-come.” Journal of Peace Research 41(3): 303–320.
de Soysa, Indra. 2002. “Paradise is a Bazaar? Greed, Creed, and Governance in Civil War, 1989–99.” Journal of Peace Research 39(4): 395–416.
de Tocqueville, Alexis, 1856/1955. The Old Régime and the French Revolution. New York: Doubleday.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper. Eck, Kristine, and Lisa Hultman, 2007. “One-Sided Violence against Civilians in War: In-
sights from New Fatality Data.” Journal of Peace Research 44(2): 233–246. Elbadawi, Ibrahim, and Nicholas Sambanis. 2000. “Why Are There So Many Conflicts in
Africa.” Journal of African Economies 9(3): 244–269. Elbadawi, Ibrahim A., and Nicholas Sambanis. 2002. “How Much War Will We See? Ex-
plaining the Prevalence of Civil War.” Journal of Conflict Resolution 46(3): 307–334. Ellingsen, Tanja. 2000. “Colorful Community or Ethnic Witches‘ Brew? Multiethnicity and
Domestic Conflict During and After the Cold War.” Journal of Conflict Resolution 44(2): 228–249.
Ellingsen, Tanja, and Nils Petter Gleditsch. 1997. “Democracy and Armed Conflict in the Third World.” In Ketil Volden, and Dan Smith (eds.), Causes of Conflict in Third World Countries. Oslo: North-South Coalition & International Peace Research Insti-tute, Oslo (69–81).
Fearon, James D. 2004. “Why Do Some Civil Wars Last So Much Longer Than Others?.” Journal of Peace Research 41(3): 275–302.
Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” Amer-ican Political Science Review 97(1): 75–90.
Freedom House, annual. Freedom in the World 2005 – The Annual Survey of Political Rights and Civil Liberties. Lanham, MD: Rowman & Littlefield. [Earlier editions from vari-ous publishers.] See also www.freedomhouse.org.
Freedom House. 2006. “Methodology”, www.freedomhouse.org/template.cfm?page=35&year=2006, accessed 5 October. Gates, Scott. 2002. “Recruitment and Allegiance: The Microfoundations of Rebellion.” Jour-
nal of Conflict Resolution 46(1): 111–130. Gates, Scott, Håvard Hegre, Mark P. Jones, and Håvard Strand. 2006. “Institutional Incon-
sistency and Political Instability: Polity Duration, 1800–2000.” American Journal of Political Science 50(4): 893–908.
30
Gates, Scott, and Håvard Strand. 2006. “Modeling the Duration of Civil Wars: Measurement and Estimation Issues,” Ch. 3 in Strand (2006).
Gleditsch, Kristian Skrede. 2002a. All International Politics Is Local. The Diffusion of Con-flict, Integration, and Democratization. Ann Arbor, MI: University of Michigan Press.
Gleditsch, Kristian Skrede. 2002b. “Expanded Trade and GDP Data.” Journal of Conflict Resolution 46(5): 712–724.
Gleditsch, Kristian Skrede. 2007. “Transnational Dimensions of Civil War.” Journal of Peace Research 44(3): 293–309.
Gleditsch, Kristian Skrede, and Michael D. Ward. 1997. “Double Take: A Re-Examination of Democracy and Autocracy in Modern Polities.” Journal of Conflict Resolution 41(3): 361–383.
Gleditsch, Kristian Skrede, and Michael D. Ward. 1999. “Interstate System Membership: A Revised List of the Independent States since 1816.” International Interactions 25(4): 393–413.
Gleditsch, Kristian Skrede, and Michael D. Ward. 2006. “The Diffusion of Democracy and the International Context of Democratization.” International Organization 60(4): 911–933.
Gleditsch, Nils Petter, and Håvard Hegre. 1997. “Peace and Democracy – Three Levels of Analysis.” Journal of Conflict Resolution 41(2): 283–310.
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and Håvard Strand. 2002. “Armed Conflict 1946–2001: A New Dataset.” Journal of Peace Research 39(5): 615–637.
Golder, Matt 2005. “Democratic Electoral Systems around the World, 1946–2000.” Electoral Studies 24(1): 103–121.
Gurr, Ted Robert. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press. Gurr, Ted Robert. 1993. “Why Minorities Rebel: A Global Analysis of Communal Mobiliza-
tion and Conflict since 1945.” International Political Science Review 14(2): 161–201. Harbom, Lotta, Stina Högbladh, and Peter Wallensteen. 2006. “Armed Conflict and Peace
Agreements.” Journal of Peace Research 43(5): 617–631. Hartzell, Caroline A., and Matthew Hoddie. 2003, “Institutionalizing Peace: Power Sharing
and Post-Civil War Conflict Management.” American Journal of Political Science 47(2): 318–332.
Hegre, Håvard. 2000. “Development and the Liberal Peace: What Does It Take to Be a Trad-ing State?” Journal of Peace Research 37(1): 5–30.
Hegre, Håvard. 2003. “Disentangling Democracy, Development as Determinants of Armed Conflict.” Paper presented at the Annual Convention of the International Studies Association, Portland, OR, 25 February–1 March, www.prio.no/files/file40692_ddcwwb.pdf.
Hegre, Håvard. 2005. Development and the Liberal Peace.” Nordic Journal of Political Econ-omy 31(1): 17–46.
Hegre, Håvard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch. 2001. “Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816–1992.” American Political Science Review 95(1): 17–33.
Hegre, Håvard, Ranveig Gissinger, and Nils Petter Gleditsch. 2003. “Globalization and In-ternal Conflict.” In Gerald Schneider, Katherine Barbieri, and Nils Petter Gleditsch, eds, Globalization and Armed Conflict. Lanham, MD: Rowman & Littlefield (251–276).
Hegre, Håvard, and Nicholas Sambanis. 2006. “Sensitivity Analysis of the Empirical Litera-ture on Civil War Onset”. Journal of Conflict Resolution 50(4): 508–535.
Henderson, Errol & J. David Singer. 2000. “Civil War in the Post-Colonial World, 1946–92.” Journal of Peace Research 37(3): 275–299.
Huntington, Samuel P. 1991. The Third Wave: Democratization in the Late Twentieth Centu-ry. Norman, OK & London: University of Oklahoma Press.
Jaggers, Keith, and Ted Robert Gurr. 1995. “Tracking Democracy’s Third Wave with the Polity III Data.” Journal of Peace Research 32(4): 469–482.
Krause, Volker, and Susumu Suzuki. 2005. “Causes of Civil War in Asia and Sub-Saharan Africa: A Comparison.” Social Science Quarterly 86(1): 160–177.
Lacina, Bethany. 2006. “Explaining the Severity of Civil War.” Journal of Conflict Resolution 50(2): 276–289.
Lacina, Bethany, and Nils Petter Gleditsch. 2005. “Monitoring Trends in Global Combat: A New Dataset of Battle Deaths.” European Journal of Population 21(2): 145–166.
31
Lichbach, Mark Irving. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of Repression and Dissent.” Journal of Conflict Resolution 31(2): 266–297.
Lichbach, Mark Irving. 1995. The Rebel's Dilemma. Ann Arbor, MI: University of Michigan Press.
Lipset, Seymour M. 1959. “Some Social Requisites of Democracy: Economic Development and Political Legitimacy.” American Political Science Review 53(1): 69–106.
Mack, Andrew. 2006. Human Security Brief. Vancouver, BC: Human Security Centre, Uni-versity of British Columbia,
www.humansecuritycentre.org/images/stories/HSBrief2006/index.html. Marshall, Monty G., and Keith Jaggers. 2003. Polity IV Project: Political Regime Characteris-
tics and Transitions, 1800–2003, www.cidcm.umd.edu/inscr/polity/index.htm. Moore, Will H. 1998. “Repression and Dissent: Substitution, Context, and Timing.” Ameri-
can Journal of Political Science 42(3): 851–873. Most, Benjamin A., and Harvey Starr. 1989. Inquiry, Logic, and International Politics. Co-
lumbia, SC: University of South Carolina Press. Mousseau, Michael. 2000. “Market Prosperity, Democratic Consolidation, and Democratic
Peace.” Journal of Conflict Resolution 44(4): 472–507. Mousseau, Michael, Håvard Hegre, and John R. Oneal. 2003. “How the Wealth of Nations
Conditions the Liberal Peace.” European Journal of International Relations 9(2): 277–314.
Muller, Edward N., and Erich Weede. 1990. “Cross-National Variations in Political Violence: A Rational Action Approach.” Journal of Conflict Resolution 34(4): 624–651.
Munck, Gerardo L., and Jay Verkuilen. 2002. “Conceptualizing and Measuring Democracy: Evaluating Alternative Indices.” Comparative Political Studies 35(1): 5–34.
Murdoch, James C., and Todd Sandler. 2002. “Economic Growth, Civil Wars, and Spatial Spillovers.” Journal of Conflict Resolution 46(1): 91–110.
Przeworski, Adam. 1991. Democracy and the Market. Cambridge: Cambridge University Press.
Przeworski, Adam, Michael E. Alvarez, José Antonio Cheibub, and Fernando Limongi. 2000. Democracy and Development. Political Institutions and Well-Being in the World, 1950–1990. Cambridge: Cambridge University Press.
Raknerud, Arvid, and Håvard Hegre, 1997. ‘The Hazard of War: Reassessing the Evidence for the Democratic Peace’, Journal of Peace Research 34(4): 385–404.
Raleigh, Clionadh, and Håvard Hegre. 2005. “Introducing ACLED: An Armed Conflict Loca-tion and Event Dataset.” paper presented at the IGCC Conference “Disaggregating the Study of Civil War and Transnational Violence.” University of California San Di-ego, La Jolla, CA, 7–8 March, www.prio.no/files/file46564_ucsd_paper_final.pdf.
Reynal-Querol, Marta. 2002. “Ethnicity, Political Systems, and Civil Wars.” Journal of Con-flict Resolution 46(1): 29–54.
Roeder, Philip G. 2001. “Ethnolinguistic Fractionalization (ELF) Indices, 1961 and 1985.” 16 February, http//:weber.ucsd.edu\~proeder\elf.htm>, accessed 15 June.
Sambanis, Nicholas. 2001. “Do Ethnic and Non-Ethnic Civil Wars Have the Same Causes? A Theoretical and Empirical Inquiry (Part 1).” Journal of Conflict Resolution 45(3): 259–282.
Sambanis, Nicholas. 2002. “A Review of Recent Advances and Future Directions in the Quantitative Literature on Civil War.” Defence and Peace Economics 13(3): 215–243.
Schneider, Gerald, and Nina Wiesehomeier. 2005. “Rules that Matter. Political Institutions and the Polarization-Conflict Nexus.” paper presented to the Summer Meeting of the “Polarization and Conflict” network, Konstanz, 2–4 June,
www.iae.csic.es/polarization/wp/PAC-Schne-Wiesehomeier.pdf. Smith, Benjamin. 2004. “Oil Wealth and Regime Survival in the Developing World, 1960–
1999”, American Journal of Political Science 48(2): 232–246. Snyder, Jack. 2000. From Voting to Violence: Democratization and Nationalist Conflict. New
York: Norton. Strand, Håvard. 2006. Reassessing the Civil Democratic Peace, PhD dissertation, Depart-
ment of Political Science, University of Oslo & Centre for the Study of Civil War, PRIO, http://www.prio.no/files/manual-import/Strand_Thesis.pdf
Strand, Håvard, 2007. “Research Design for article on Democracy and Civil War”, www.prio.no/cscw/datasets.
32
Strand, Håvard, Joachim Carlsen, Nils Petter Gleditsch, Håvard Hegre, Christin Ormhaug, and Lars Wilhelmsen. 2005. Armed Conflict Dataset Codebook, Version 3–2005. Os-lo: Centre for the Study of Civil War, PRIO, www.prio.no/cscw/armedconflict.
Treier, Shawn, and Simon Jackman. 2005. “Democracy as a Latent Variable.” Department of Political Science, Stanford University, http://polmeth.wustl.edu/retrieve.php?id=562.
Urdal, Henrik. 2005. “People vs. Malthus: Population Pressure, Environmental Degradation and Armed Conflict Revisited.” Journal of Peace Research 42(4): 417–434.
Vanhanen, Tatu. 2000. “A New Dataset for Measuring Democracy, 1810–1998.” Journal of Peace Research 37(2): 251–265. Data available at www.prio.no/jpr/datasets.
Vreeland, James Raymond. 2008. “The Effect of Political Regime on Civil War: Unpacking Anocracy.” Journal of Conflict Resolution 52(3), in press,
Wilkinson, Stephen I. 2004. Votes and Violence: Electoral Competition and Ethnic Riots in India. Cambridge: Cambridge University Press.
Zanger, Sabine C. 2000. “A Global Analysis of the Effect of Political Regime Changes on Life Integrity Violations, 1977–93.” Journal of Peace Research 37(2): 213–233.
33
Table I. Democracy and Four Measures of Civil War, 1950–2000
Onset Incidence Duration Severity
Polity IV democracy-autocracy index
0.60 ** 0.95 3.61 *** 0.37 ***
Polity IV democracy-autocracy index squared
0.22 *** 0.30 *** 1.40 0.37 ***
SIP democracy index 0.53 *** 0.90 2.80 ** 0.41 *** SIP democracy index squared 0.41 *** 0.70 *** 3.66 ** 0.44 *** Vanhanen index of democracy 0.24 *** 0.49 *** 4.06 0.21 *** Vanhanen ID, squared 0.005 *** 0.072 *** 1.81 0.059 *** Freedom House index 0.34 *** 0.38 *** 2.06 0.28 *** Freedom House index squared 0.44 *** 0.51 *** 0.64 1.19 Proximity of regime change 2.61 *** 2.17 *** 0.76 2.02 *** Proximity of election 1.15 0.88 1.70 0.75 ** Difference to political system in neighborhood
0.75 1.29 ** 3.78 *** 0.52 ***
Difference squared 1.84 2.77 *** 1.22 0.17 *** Presidentialism 0.96 1.44 *** 0.50 1.79 *** Proportional representation 0.32 *** 0.70 *** 2.66 0.76 * Federalism 2.56 *** 0.69 ** 1.35 1.08 The figures for the onset and incidence columns are relative risk ratios; the figures for duration are time ratios; and the figures for severity are casualty ratios. The estimates are based on a dataset with 159 countries and 204 civil wars (corresponding to 5,848 country-years). Within each column, all analyses are based on the same set of countries and conflicts. The exceptions are in the final three columns, which are restricted to 49 wars in 93 democracies (2,366 country-years). For all analyses, we removed observations with missing information for the control variables to make the results im-mediately comparable to those reported in Table IV. *** p<0.01; ** p<0.05; * p<0.1. Here and in the subsequent tables we report two-sided tests of the hypothesis that the exponent of the coefficient is not equal to one. Se Appendix with the replication data for the details and also Strand (2007).
34
Table II. The inverted U, Controlling for Four Single Control Variables, Onset of Civil War, 1950–2000
Democracy Democra-
cy squared
Squared difference to political system in neighborhood
Proximity to elections
Bivariate results (159/204)
0.53 *** 0.41 *** 0.75 1.15
Control for political sta-bility
0.59 *** 0.52 ** 2.42 ** 1.67
High level of development 0.30 ** 0.08 *** 4.54 4.54 Low level of development 0.92 0.76 1.99 * 1.91 Governmental conflict 0.41 *** 0.30 *** 0.74 2.09 Territorial conflict 0.85 0.58 8.40 *** 2.56 Cold War 0.79 0.60 1.57 2.66 * Post-Cold War
0.34 *** 0.24 *** 5.03 *** 1.94
All results, except the first line (which is copied from Table I) are based on trivariate models of onset. All figures are relative risk ratios estimated using calendar-time Cox regression. The estimates are based on a dataset with 159 countries and 204 civil wars. Within each column, all analyses are based on the same set of countries and conflicts. In the analyses in the last six lines this dataset are divided into three pairs of sub-samples. The high-development subset has 25 wars in 80 countries The gov-ernmental subset has 119 wars in 159 countries. The Cold War subset has 122 wars in 138 coun-tries. Democracy is measured by the SIP index. *** p<0.01; ** p<0.05; * p<0.1, based on two-sided tests of the hypothesis that the exponent of the coefficient is not equal to one.
35
Table III. Democracy and Civil War, with Control Variables, 1950–2000
Onset Incidence Duration Severity
with Polity Democracy 0.90 0.82 2.06 0.40 *** Democracy squared 0.42 *** 0.53 *** 0.98 0.52 *** with SIP Democracy 0.80 0.73 1.23 0.55 *** Democracy squared 0.58 * 0.79 2.29 0.95 Proximity of regime change 1.35 1.06 1.10 1.63 *** Proximity of election 1.58 1.13 1.95 0.88 Difference to political system in neighborhood
1.26 1.49 1.93 1.41
Difference squared 2.38 * 2.60 ** 0.12 0.29 *** Presidentialism 0.57 0.62 0.48 2.40 *** Proportional representation 1.002 2.39 ** 10.55 ** 0.48 *** Federalism 0.57 0.32 ** 0.25 2.67 ** The figures in the onset and incidence column are relative risk ratios, the figures for duration are time ratios, and the figures for severity are casualty ratios. The full set of control variables (results not reported in the table) is found in Table IV and described in more detail in Strand (2007). The esti-mates are based on a dataset with 159 countries and 204 civil wars (corresponding to 5,848 country-years). Within each column, all analyses are based on the same set of countries and conflicts. The exceptions are in the final three columns, which are restricted to 49 wars in 93 democracies (2,366 country-years). *** p<0.01; ** p<0.05; * p<0.1, based on two-sided tests of the hypothesis that the exponent of the coefficient not being equal to one.
36
Table IV. The Correlates of Internal Armed Conflict Onset, Hazard Ratios. Multivariate Results with Control Variables, 1950–2000
Model 1 Model 2 Model 3 Model 4
SIP 0.802 1.115 1.207 0.719
(-0.581) (0.415) (0.473) (-0.619)
SIP squared 0.576* 0.642 0.520** 0.509*
(-1.880) (-1.551) (-1.975) (-1.784)
Proximity to regime change 1.351 1.590* 1.730* 1.699
(1.039) (1.696) (1.740) (1.614)
Proximity to independence 2.792** 2.909** 1.419
(2.312) (1.981) (0.533)
Proximity to first real election 1.577 1.157 3.035**
(1.077) (0.345) (2.391)
Proximity to other elections 0.925 0.983 1.119
(-0.230) (0.015) (0.224)
Democratic difference 1.262 1.030 1.046
(0.636) (0.085) (0.098)
Democratic difference squared 2.377* 2.060 1.676
(1.711) (1.110) (0.693)
Average income (ln) 0.664*** 0.637*** 0.622*** 0.686**
(-3.488) (-3.948) (-3.603) (-2.564)
Population (ln) 1.340*** 1.325*** 1.287*** 1.232***
(4.898) (4.586) (3.322) (3.267)
Proximity to conflict 3.678*** 3.465*** 1.752*** 7.163***
(6.308) (5.948) (2.287) (7.579)
Ethno-linguistic fractionalization 2.220** 2.327** 2.087* 1.734
(2.466) (2.558) (1.930) (1.458)
Ethno-linguistic fractionalization squared 0.046** 0.044** 0.099* 0.031**
(-2.435) (-2.493) (-1.718) (-2.126)
Oil exporter 1.665** 1.706** 1.770** 1.800*
(2.005) (2.089) (2.182) (1.761)
Log likelihood -875.98 -881.07 -695.17 -501.46
Log likelihood null model -983.92 -983.92 -754.29 -579.66
N 25,972 25,972 19,537 14,801
Number of countries 159 159 159 159
Number of civil wars 204 204 157 121
Model 3 excludes intermittent conflicts. Model 4 includes only conflicts with more than 1,000 battle deaths. The figures reported are relative risk ratios, with z-scores in parentheses. N refers to the number of snapshots analyzed. At each of the 204 onsets, all countries independent at that moment are observed. Since a number of the 159 countries included in our study gained their independence after 1945, our N is lower than 31,959 (204 * 159). *** p<0.01; ** p<0.05; * p<0.1, based on two-sided tests of the hypothesis that the exponent of the coefficient is not equal to one.
37
Table V. Correlates of the Severity of Internal Armed Conflict, Ratios. Multi-variate Results with Control Variables, 1950–2000 Model 1 Model 2 Lagged dependent variable (ln) 1.38*** 1.39*** (19.22) (19.46) Polity IV 0.40*** (–3.53) Polity IV squared 0.52*** (–3.34) SIP democracy index 0.55*** (–2.29) SIP democracy index squared 0.95 (–0.28) Difference to political system in neighborhood 1.76** 1.41 (2.41) (1.37) Difference squared 0.39*** 0.29*** (–2.69) (–3.39) Proximity to regime change 1.33* 1.63*** (1.85) (3.16) Average income (ln) 0.85*** 0.84*** (–3.30) (–3.29) Population (ln) 1.04 1.03 (0.87) (0.67) Election year 0.88 0.88 (–1.02) (–0,97) Ethno-linguistic fractionalization 1.17 1.34 (1.28) (1.28) Ethno-linguistic fractionalization squared 0.033*** 0.042*** (–4.15) (–3.76) Oil exporter (>1/3 of total exports) 1.35** 1.40** (2.09) (2.30) Log likelihood –1,547.39 –1,595.93 Log likelihood null model –1,771.30 –1,817.57 Number of country-years 877 897 The figures reported are exponentiated coefficients, to be interpreted as ratios relative to the baseline, with t-scores in parentheses. *** p<0.01; ** p<0.05; * p<0.1, based on two-sided tests of the hypothesis that the exponent of the coefficient is not equal to one.
38
Figure 1. Regime Type and the Severity of Civil War, 1989–2004
Conflict data from the Uppsala/PRIO conflict data (Gleditsch et al. 2002), battle deaths from Lacina and Gleditsch (2005), and regime type data from Gates, Hegre, Jones, and Strand ( 2006) with some additions for countries with missing scores in their dataset. Countries coded as democracies and au-tocracies have had stable regime types (democracy = SIP index > 0.80, autocracy = SIP < 0.25). All others – including politically unstable countries and countries stably located in the middle of the scale – are coded as semi-democracies. For the distinction between conflicts over government and territory, see Buhaug (2006). The dots and asterisks are placed at the center of the conflict zone (Buhaug and Gates 2002). The larger the dot or asterisk, the greater the cumulative number of battle deaths over this 16-year period. Only conflicts active during this period have been included. The definitions and time span used in this figure differs marginally from that used in the analysis due for practical rea-sons.
39
Appendix: Democracies with internal armed conflict, 1950–
2000
In this article, we use the Uppsala/PRIO dataset to test the hypotheses about civil
war and democracy. This dataset is updated annually and published in Journal of
Peace Research (no. 5), and posted on www.prio.no/cscw/armedconflict. In this
article we use the conflict data for the period 1950–2000. For the severity of con-
flict, we use the Lacina battle deaths dataset, posted on
www.prio.no/cscw/cross/battledeaths cf. Lacina & Gleditsch (2005).
During the period 1950–2000 there were 199 internal armed conflicts. This
figure assumes that when a conflict has been inactive (i.e. has had fewer than 25
battle deaths) for more than 10 years or is conducted by new actors, we define it as
a new conflict.
Below, we list the 52 internal armed conflicts during this period which partly
or completely occurred in a democracy. The conflicts are labeled with the name of
the country and in the case of territorial conflicts with the name of the territory.
The information provided is their start and end dates; the duration (measured in
days); the percentage of these days fought under a democratic form of governance;
the regime type at the onset of the conflict (autocratic, semi-democratic, or demo-
cratic); and finally the best estimate for the total number of battle-related fatalities
which occurred during a democratic form of governance.
Country Incompatibility Start End Duration % demo-
cratic
Regime type
at onset
Severity
Argentina Government 01 Mar 73 31 Dec 77 1,767 62.8 % Aut 1, 361
Bangladesh Terr: Chittagong Hill
Tracts
15 Aug 75 02 Dec 97 8, 146 27.7 % Semi-dem 175
Burma Government 28 Mar 48 31 Dec 94 15, 959 13.4 % Semi-dem 1, 284
Burma Terr: Arakan 04 Jan 48 31 Dec 94 15, 968 13.4 % Semi-dem 146
Burma Terr: Kachin 31 Dec 61 31 Dec 92 11, 324 0.5 % Dem 1, 164
Burma Terr: Karen 31 Jan 49 31 Dec 00 18, 018 11.9 % Semi-dem 1, 284
Burma Terr: Karenni 01 Jan 57 31 Dec 57 365 100.0 % Dem 290
Burma Terr: Mon 04 Jan 48 15 Nov 63 5, 795 36.9 % Semi-dem 321
Burma Terr: Shan 30 Nov 59 31 Dec 00 11, 617 7.1 % Dem 241
Colombia Government 16 Aug 66 31 Dec 00 12, 557 76.8 % Semi-dem 19, 779
Congo Government 03 Nov 93 16 Nov 99 985 22.5 % Dem 5, 660
El Salvador Government 01 Oct 79 31 Dec 91 4, 475 61.9 % Semi-dem 24, 392
Gambia Government 30 Jul 81 06 Aug 81 8 100.0 % Dem 650
Guatemala Government 01 Oct 66 31 Dec 00 12, 511 14.6 % Semi-dem 125
India Government 25 May 67 19 Jul 72 1, 883 100.0 % Dem 300
India Government 10 Oct 90 31 Dec 00 3, 736 100.0 % Dem 1, 658
India Terr: Assam 29 May 90 31 Dec 00 3, 125 100.0 % Dem 1, 905
India Terr: Bodoland 16 Mar 89 31 Dec 00 3, 578 100.0 % Dem 250
India Terr: Kashmir 11 Dec 89 31 Dec 00 4, 039 100.0 % Dem 18, 360
India Terr: Manipur 31 Jul 82 31 Dec 00 5, 634 100.0 % Dem 925
India Terr: Mizoram 01 Sep 66 31 Oct 68 792 100.0 % Dem 1, 500
India Terr: Nagaland 01 Jan 56 30 Jun 68 4, 565 100.0 % Dem 2, 078
India Terr: Nagaland 31 Jul 92 31 Dec 00 2, 184 100.0 % Dem 245
India Terr: Pun-
jab/Khalistan
20 Aug 83 12 Sep 93 3, 677 100.0 % Dem 18, 875
India Terr: Tripura 01 Jan 78 31 Dec 00 6, 880 100.0 % Dem 1, 097
Indonesia Terr: Aceh 08 Sep 90 31 Dec 00 1, 027 42.6 % Aut 349
Israel Terr: Palestine 01 Jan 49 31 Dec 00 18, 993 99.9 % Semi-dem 13, 175
40
Lesotho Government 23 Sep 98 14 Oct 98 22 100.0 % Dem 107
Malaysia Government 01 Jan 58 31 Jul 60 943 36.9 % Semi-dem 22
Malaysia Terr: North Borneo 01 Jan 63 11 Aug 66 1, 319 100.0 % Dem 400
Niger Terr: Air and
Azawad
01 Oct 92 27 Nov 97 910 80.8 % Transitional 59
Pakistan Government 01 Jun 90 31 Mar 96 670 100.0 % Dem 2, 525
Pakistan Terr: Baluchistan 01 Jan 74 05 Jul 77 1, 282 100.0 % Dem 8, 332
Papua New Guinea Terr: Bougainville 01 Dec 89 31 Dec 96 2, 588 100.0 % Dem 375
Peru Government 01 Oct 65 31 Jan 66 123 100.0 % Dem 138
Peru Government 22 Aug 82 31 Dec 99 6, 341 55.4 % Dem 23, 433
Philippines Government 04 Jul 46 17 May 54 2, 875 49.3 % Semi-dem 4, 504
Philippines Government 21 Sep 72 31 Dec 00 10, 329 49.2 % Semi-dem 7, 619
Philippines Terr: Mindanao 20 Aug 70 31 Dec 00 9, 996 39.9 % Semi-dem 5, 814
Senegal Terr: Casamance 01 Jun 90 31 Dec 00 3, 867 7.4 % Semi-dem 39
Spain Terr: Basque 01 Jan 80 31 Dec 92 1, 827 100.0 % Dem 245
Sri Lanka Government 30 Apr 71 09 Jun 71 41 100.0 % Dem 1, 630
Sri Lanka Government 01 Feb 89 28 Feb 90 393 100.0 % Dem 5, 025
Sri Lanka Terr: Eelam 01 Jul 83 31 Dec 00 6, 394 100.0 % Dem 53, 975
Sudan Government 01 Jan 63 31 Jan 72 3, 318 11.7 % Aut 2, 395
Sudan Government 16 May 83 31 Dec 00 6, 440 18.4 % Aut 12, 975
Trinidad and Tobago Government 27 Jul 90 01 Aug 90 6 100.0 % Dem 30
Turkey Government 13 Jul 91 31 Oct 92 477 100.0 % Dem 50
Turkey Terr: Kurdistan 15 Aug 84 31 Dec 00 5, 983 68.0 % Semi-dem 33, 080
United Kingdom Terr: Northern
Ireland
01 Jan 71 15 Aug 98 7, 671 100.0 % Dem 3, 149
Venezuela Government 02 Jun 62 03 Jun 62 2 100.0 % Dem 400
Venezuela Government 04 Feb 92 29 Nov 92 300 100.0 % Dem 183